Comparing L1000 and Cell painting compound's values/scores based on their replicates

• Visualizing the median scores of cell painting cpds vs L1000 cpds using scatter plot.

• Visualizing the median scores distribution of both cell painting and L1000.

• Visualizing the signature strength of Cell painting vs L1000 compounds using scatter plot.

• Visualizing the MAS (Morphological activity score) of Cell painting vs TAS (Transcriptional activity score) L1000 compounds using scatter plot.

import os
import pathlib
import pandas as pd
import numpy as np
from collections import defaultdict
import matplotlib.pyplot as plt
%matplotlib inline
import seaborn as sns
import pickle
sns.set_style("darkgrid")
##sns.set_palette(["red", "green", "orange","blue","gray","purple"])
sns.set_context("talk")

cp_level4_path = '../cell_painting/cellpainting_lvl4_cpd_replicate_datasets'
L1000_level4_path = '../L1000/L1000_lvl4_cpd_replicate_datasets'

df_cp_med = pd.read_csv(os.path.join(cp_level4_path, 'cpd_replicate_median_scores.csv'))
df_L1_med = pd.read_csv(os.path.join(L1000_level4_path, 'cpd_replicate_median_scores.csv'))

df_cp_pvals = pd.read_csv(os.path.join(cp_level4_path, 'cpd_replicate_p_values.csv'))
df_L1_pvals = pd.read_csv(os.path.join(L1000_level4_path, 'cpd_replicate_p_values.csv'))

df_cp_all = pd.read_csv(os.path.join(cp_level4_path, 'cp_all_scores.csv'))
df_L1_all = pd.read_csv(os.path.join(L1000_level4_path, 'L1000_all_scores.csv'))

with open(os.path.join(cp_level4_path, 'CP_dmso_95_percentile_MAS.pickle'), 'rb') as handle:
    cp_dmso_95pct = pickle.load(handle)
with open(os.path.join(L1000_level4_path, 'L1000_dmso_95_percentile_TAS.pickle'), 'rb') as handle:
    L1_dmso_95pct = pickle.load(handle)

with open(os.path.join(cp_level4_path, 'null_dist_medians_per_dose.pickle'), 'rb') as handle:
    null_dist_med_cp = pickle.load(handle)
    
with open(os.path.join(L1000_level4_path, 'null_dist_medians_per_dose.pickle'), 'rb') as handle:
    null_dist_med_L1000 = pickle.load(handle)

def reproducible_dose(df, df_med):
    """
    This function computes how many reproducible median correlation scores 
    each compound has across all doses, (out of the 6 doses based on p-values)
    """
    df_new = df.set_index('cpd').rename_axis(None, axis=0).drop(['no_of_replicates'], axis = 1).copy()
    cpd_values = {cpd:sum(df_new.loc[cpd] <= 0.05) for cpd in df_new.index}
    df['No_of_reproducible_doses'] = cpd_values.values()
    df_med['No_of_reproducible_doses'] = cpd_values.values()
    
    return df, df_med

df_cp_pvals, df_cp_med = reproducible_dose(df_cp_pvals, df_cp_med)
df_L1_pvals, df_L1_med = reproducible_dose(df_L1_pvals, df_L1_med)

def rename_cols(df):
    'Rename columns from dose number to actual doses'
    
    df.rename(columns= {'dose_1' : '0.04 uM', 'dose_2':'0.12 uM', 'dose_3':'0.37 uM',
                        'dose_4': '1.11 uM', 'dose_5':'3.33 uM', 'dose_6':'10 uM'}, inplace = True)
    return df

df_cp_med = rename_cols(df_cp_med)
df_L1_med = rename_cols(df_L1_med)

df_cp_pvals = rename_cols(df_cp_pvals)
df_L1_pvals = rename_cols(df_L1_pvals)

def melt_df(df, col_name):
    """Reconstruct the dataframe into 4 columns: cpd, cpd_size, No_of_reproducible_doses, doses"""
    df_melt = df.melt(id_vars=['cpd', 'no_of_replicates', 'No_of_reproducible_doses'], var_name="dose", value_name=col_name)
    return df_melt

def melt_L1000_CP(df_cp,df_L1):
    """
    Reconstruct Cell painting and L1000 dataframes, 
    by tranforming the dose columns into vertical rows
    """
    cp_cpd = df_cp.cpd.values.tolist()
    L1_cpd = df_L1.cpd.values.tolist()
    all_cpds = [x for x in cp_cpd if x in L1_cpd]
    df_cp_med =df_cp.loc[df_cp['cpd'].isin(all_cpds)].reset_index(drop=True)
    df_L1_med = df_L1.loc[df_L1['cpd'].isin(all_cpds)].reset_index(drop=True)
    df_cp_melt = melt_df(df_cp_med, 'Cell painting')
    df_L1_melt = melt_df(df_L1_med,'L1000')
    
    return df_cp_melt, df_L1_melt

df_cp_melt, df_L1_melt = melt_L1000_CP(df_cp_med,df_L1_med)

len(df_L1_melt['cpd'].unique())

1327

df_L1_melt.head()

	cpd	no_of_replicates	No_of_reproducible_doses	dose	L1000
0	17-hydroxyprogesterone-caproate	3	3	0.04 uM	0.024738
1	2-iminobiotin	2	1	0.04 uM	-0.133693
2	3-amino-benzamide	3	1	0.04 uM	0.051959
3	3-deazaadenosine	2	0	0.04 uM	0.065967
4	ABT-737	3	1	0.04 uM	0.016400

def merge_L1000_CP(df_cp, df_L1):
    
    """Merge L1000 and Cell painting dataframes based on compounds and doses found in both dataframes"""
    
    df_cp_new = df_cp.rename(columns={"Cell painting": "median scores"})
    df_cp_new['profile_tech'] = 'Cell painting'
    df_L1_new = df_L1.rename(columns={"L1000": "median scores"})
    df_L1_new['profile_tech'] = 'L1000'
    df_cp_L1 = pd.concat([df_cp_new, df_L1_new], ignore_index=True)
    
    df_cp_nw = df_cp.rename(columns={'no_of_replicates': "no_of_replicates_CP", 
                          "No_of_reproducible_doses": "No_of_reproducible_doses-Cell painting"})
    df_L1_nw = df_L1.rename(columns={'no_of_replicates': "no_of_replicates_L1000",
                          "No_of_reproducible_doses": "No_of_reproducible_doses-L1000"})
    df_L1_cp_merge = df_cp_nw.merge(df_L1_nw, on=['cpd', 'dose'])
    return df_cp_L1, df_L1_cp_merge

df_cp_L1_melt, df_L1_cp_med = merge_L1000_CP(df_cp_melt, df_L1_melt)

df_cp_L1_melt.head()

	cpd	no_of_replicates	No_of_reproducible_doses	dose	median scores	profile_tech
0	17-hydroxyprogesterone-caproate	5	6	0.04 uM	0.060655	Cell painting
1	2-iminobiotin	5	1	0.04 uM	0.042700	Cell painting
2	3-amino-benzamide	5	5	0.04 uM	0.104273	Cell painting
3	3-deazaadenosine	5	3	0.04 uM	0.010217	Cell painting
4	ABT-737	5	5	0.04 uM	0.004470	Cell painting

cp_95pct = [np.percentile(null_dist_med_cp[keys],95) for keys in null_dist_med_cp]

L1000_95pct = [np.percentile(null_dist_med_L1000[keys],95) for keys in null_dist_med_L1000]

- Plot median scores of L1000 vs Cell painting compound replicates

def plot_median_scores(df, L1000_95pct, cp_95pct, title, path, file_name, hue = 'Cell painting', 
                       alp =0.4, size =(50,300)):
    
    """Plot median scores of L1000 versus Cell painting"""
    
    if not os.path.exists(path):
        os.mkdir(path)
    rel_plt = sns.relplot(data=df, x="Cell painting", y="L1000", col="dose", 
                          hue = 'No_of_reproducible_doses-'+ hue, size = 'No_of_reproducible_doses-'+ hue, sizes= size,
                          kind="scatter", palette=sns.color_palette("crest", as_cmap=True), 
                          col_wrap=3, height=5.5, aspect=1.3, alpha = alp)
    rel_plt.fig.suptitle(title)
    rel_plt.fig.subplots_adjust(top=.91)
    for idx, val in enumerate(zip(L1000_95pct, cp_95pct)):
        rel_plt.axes[idx].axhline(y = val[0], ls='--', color = 'black', alpha = 0.7, label = '95th null distribution')
        rel_plt.axes[idx].axvline(x = val[1], ls='--', color = 'black', alpha = 0.7)
    rel_leg = rel_plt._legend
    rel_leg.set_bbox_to_anchor([0.90, 0.51])
    plt.legend(bbox_to_anchor=(1.01, 0.505), loc='lower left')
    rel_plt.legend.set_title('Number of\nreproducible\ndoses\n' + '('+ hue +')')
    plt.savefig(os.path.join(path, file_name))
    plt.show()

plot_median_scores(df_L1_cp_med, L1000_95pct, cp_95pct, 
                   "L1000 vs Cell Painting median correlation scores for compound replicates across all doses",
                   'L1000_cp_figures', 'median_scores_plot_1a.png')

plot_median_scores(df_L1_cp_med, L1000_95pct, cp_95pct, 
                   "L1000 vs Cell Painting median correlation scores for compound replicates across all doses",
                   'L1000_cp_figures', 'median_scores_plot_1b.png', hue = 'L1000')

df_L1_cp_med.head()

	cpd	no_of_replicates_CP	No_of_reproducible_doses-Cell painting	dose	Cell painting	no_of_replicates_L1000	No_of_reproducible_doses-L1000	L1000
0	17-hydroxyprogesterone-caproate	5	6	0.04 uM	0.060655	3	3	0.024738
1	2-iminobiotin	5	1	0.04 uM	0.042700	2	1	-0.133693
2	3-amino-benzamide	5	5	0.04 uM	0.104273	3	1	0.051959
3	3-deazaadenosine	5	3	0.04 uM	0.010217	2	0	0.065967
4	ABT-737	5	5	0.04 uM	0.004470	3	1	0.016400

df_highcp_lowL1_med = df_L1_cp_med[(df_L1_cp_med['L1000'] < 0.2) & \
                                   (df_L1_cp_med['Cell painting'] > 0.8)].reset_index(drop=True).copy()

df_highcp_lowL1_med

	cpd	no_of_replicates_CP	No_of_reproducible_doses-Cell painting	dose	Cell painting	no_of_replicates_L1000	No_of_reproducible_doses-L1000	L1000
0	clofarabine	5	6	0.37 uM	0.802077	3	4	0.144169
1	amiodarone	5	4	10 uM	0.839273	3	3	0.184575

# plot_median_scores(df_highcp_lowL1_med, L1000_95pct, cp_95pct, 
#                    "Low L1000 vs High Cell Painting median correlation scores for compound replicates",
#                    'L1000_cp_figures', 'lowL1000_HighCP_med_scores1b.png', hue = 'Cell painting', alp =0.7)

# plot_median_scores(df_highcp_lowL1_med, L1000_95pct, cp_95pct, 
#                    "Low L1000 vs High Cell Painting median correlation scores for compound replicates",
#                    'L1000_cp_figures', 'lowL1000_HighCP_med_scores1b.png', hue = 'L1000', alp =0.7)

df_lowcp_highL1_med = df_L1_cp_med[(df_L1_cp_med['L1000'] > 0.4) & \
                                   (df_L1_cp_med['Cell painting'] < 0.2)].reset_index(drop=True).copy()

df_lowcp_highL1_med

	cpd	no_of_replicates_CP	No_of_reproducible_doses-Cell painting	dose	Cell painting	no_of_replicates_L1000	No_of_reproducible_doses-L1000	L1000
0	RHO-kinase-inhibitor-III[rockout]	5	6	0.04 uM	0.112166	2	3	0.581414
1	alvocidib	5	6	0.04 uM	0.153591	3	6	0.417900
2	lasalocid	4	6	0.04 uM	0.075003	3	6	0.623312
3	l-ergothioneine	5	6	0.12 uM	0.058396	2	5	0.640835
4	lasalocid	4	6	0.12 uM	0.076381	3	6	0.469924
5	rupatadine	5	4	0.12 uM	0.027660	3	4	0.400698
6	l-ergothioneine	5	6	0.37 uM	0.063974	2	5	0.569795
7	acrivastine	5	3	1.11 uM	0.029043	3	4	0.432599
8	amg-517	5	5	1.11 uM	0.059216	3	4	0.620393
9	enzastaurin	5	4	1.11 uM	0.095641	2	4	0.504250
10	gliclazide	5	5	1.11 uM	0.176416	3	5	0.459611
11	iniparib	5	5	1.11 uM	0.047734	3	6	0.461702
12	l-ergothioneine	5	6	1.11 uM	0.049568	2	5	0.598063
13	macitentan	5	6	1.11 uM	0.091236	2	4	0.445253
14	tetrahydrouridine	5	5	1.11 uM	0.045648	2	3	0.742995
15	l-ergothioneine	5	6	3.33 uM	0.060725	2	5	0.542090
16	rupatadine	5	4	3.33 uM	0.064906	3	4	0.433176
17	tofacitinib	5	6	3.33 uM	0.070028	2	3	0.585557
18	zd-7114	5	6	10 uM	0.092886	2	1	0.470734

plot_median_scores(df_lowcp_highL1_med, L1000_95pct, cp_95pct, 
                   "High L1000 vs Low Cell Painting median scores for compound replicates",
                   'L1000_cp_figures', 'HighL1000_LowCP_med_scores1b.png', hue = 'L1000')

def plot_median_distribution(df, title, path, file_name, plot_type = "hist"):
    """
    This function plot the frequency distribution of the median scores 
    of compounds measured in cell painting and L1000
    """
    if not os.path.exists(path):
        os.mkdir(path)
    dist_plt = sns.displot(df, x = 'median scores', col = 'dose', col_wrap=3, hue='profile_tech',
                          palette = sns.set_palette(["blue","red"]), height=5.6, aspect=1.3, kind=plot_type, alpha=0.4)
    dist_plt.legend.set_title("")
    dist_plt.fig.suptitle(title)
    dist_plt.fig.subplots_adjust(top=.90)
    plt.savefig(os.path.join(path, file_name))
    plt.show()

plot_median_distribution(df_cp_L1_melt, 'Distribution of median correlation scores across L1000 and Cell painting data per doses', 
                         'L1000_cp_figures', 'median_score_distribution.png', plot_type = "hist")

def plot_median_dist(df, path, file_name):
    
    """
    This function plot the frequency distribution of the median scores 
    of compounds measured in cell painting and L1000
    """
    if not os.path.exists(path):
        os.mkdir(path)
        
    dist = sns.displot(df, x="median scores", col="profile_tech", col_wrap=2, hue = 'dose', 
                       palette = sns.set_palette(["red", "green", "orange","blue","gray","purple"]), 
                       height=6.5, aspect=1.3, multiple="stack")
    dist.fig.suptitle('Distribution of median correlation scores across L1000 and Cell painting data per doses')
    dist.fig.subplots_adjust(top=.85)
    axes = dist.axes.flatten()
    axes[0].set_title("Cell Painting")
    axes[1].set_title("L1000")
    plt.savefig(os.path.join(path, file_name))
    plt.show()

plot_median_dist(df_cp_L1_melt, 'L1000_cp_figures', 'median_score_dist.png')

- Compounds with reproducible median correlation values (i.e. p_values below 0.05) in 1.11 - 10uM doses

def stat_significant_df(df_cp_pvalues, df_L1_pvalues, df_cp_medscores, df_L1_medscores):
    """
    This function compounds with reproducible median scores 
    across the last 3 doses (1.11-10uM) found in both cell painting and L1000
    """
    stat_cp_cpds = df_cp_pvalues[df_cp_pvalues['No_of_reproducible_doses'] >= 4]['cpd'].values.tolist()
    stat_L1_cpds = df_L1_pvalues[df_L1_pvalues['No_of_reproducible_doses'] >= 4]['cpd'].values.tolist()
    stat_cpds = [x for x in stat_cp_cpds if x in stat_L1_cpds]
    df_stat_cp = df_cp_medscores.loc[df_cp_medscores['cpd'].isin(stat_cpds)].reset_index(drop=True)
    df_stat_L1 = df_L1_medscores.loc[df_L1_medscores['cpd'].isin(stat_cpds)].reset_index(drop=True)
    
    return df_stat_cp, df_stat_L1

df_stat_cp, df_stat_L1 = stat_significant_df(df_cp_pvals, df_L1_pvals, df_cp_melt, df_L1_melt)

df_cp_L1_stat_melt, df_L1_cp_stat_med = merge_L1000_CP(df_stat_cp, df_stat_L1)

df_L1_cp_stat_med.head()

	cpd	no_of_replicates_CP	No_of_reproducible_doses-Cell painting	dose	Cell painting	no_of_replicates_L1000	No_of_reproducible_doses-L1000	L1000
0	AT-7519	5	6	0.04 uM	0.143624	3	4	0.123918
1	AZD-5438	5	6	0.04 uM	0.080211	3	6	0.258015
2	AZD-8330	5	6	0.04 uM	0.529970	3	6	0.289763
3	BI-2536	5	6	0.04 uM	0.705653	3	6	0.368031
4	BIBX-1382	5	5	0.04 uM	0.074880	3	4	0.059230

plot_median_scores(df_L1_cp_stat_med, L1000_95pct, cp_95pct, 
                   "L1000 vs Cell painting reproducible compound median correlation scores across all doses",
                   'L1000_cp_figures', 'stat_sign_med_scores_plot1a.png',  alp =0.6, size =(100,300))

plot_median_scores(df_L1_cp_stat_med, L1000_95pct, cp_95pct, 
                   "L1000 vs Cell painting reproducible compound median correlation scores across all doses",
                   'L1000_cp_figures', 'stat_sign_med_scores_plot1b.png',
                   hue = 'L1000', alp =0.6, size =(100,300))

plot_median_distribution(df_cp_L1_stat_melt, 'Distribution of reproducible median correlation scores across L1000 and Cell painting', 
                         'L1000_cp_figures', 'stat_med_score_dist.png', plot_type = "hist")

- Number of compounds with reproducible median scores for Cell painting and L1000 across all doses

def reproducible_no_of_cpd(df_L1, df_cp):
    """
    This function returns a dataframe that includes number of compounds 
    with reproducible median scores in each of the 6 doses for both cell painting and L1000
    """
    columns = ['0.04 uM','0.12 uM','0.37 uM','1.11 uM','3.33 uM','10 uM']
    no_of_L1_cpd = {col:df_L1.loc[(df_L1[col] <= 0.05)].shape[0] for col in columns}
    no_of_cp_cpd = {col:df_cp.loc[(df_cp[col] <= 0.05)].shape[0] for col in columns}
    df_cpd = pd.DataFrame.from_dict(no_of_L1_cpd, orient='index', columns = ['L1000'])
    df_cpd['Cell painting'] = no_of_cp_cpd.values()
    
    return df_cpd

df_cpd = reproducible_no_of_cpd(df_L1_pvals, df_cp_pvals)

def plot_no_of_cpd_dose(df, path, file_name):
    
    """Plot number of compounds with reproducible median scores for Cell painting and L1000 per dose"""
    if not os.path.exists(path):
        os.mkdir(path)   
    fig, axes = plt.subplots(2, 3, sharey=True)
    fig.suptitle('Number of compounds with reproducible median correlation scores')
    fig.subplots_adjust(top=.91)
    axes = axes.flatten()
    fig.set_size_inches(22.7, 12.27)
    columns = ['0.04 uM','0.12 uM','0.37 uM','1.11 uM','3.33 uM','10 uM']
    for idx, col in enumerate(columns):
        sns.barplot(ax=axes[idx], x=df.loc[col].index, y=df.loc[col].values, palette=['#eb3434', '#eb7a34'])
        axes[idx].set_title(df.loc[col].name)
    [axes[i].set_ylabel("Number of compounds") for i in range(0, 1)]
    [axes[i].set_ylabel("Number of compounds") for i in range(3, 4)] 
    plt.savefig(os.path.join(path, file_name))
    plt.show()

plot_no_of_cpd_dose(df_cpd, 'L1000_cp_figures', 'no_of_reprod_cpd_per_dose.png')

def plot_no_of_cpd(df, path, file_name):
    
    """Plot number of compounds with reproducible median correlation scores for Cell painting and L1000"""
    if not os.path.exists(path):
        os.mkdir(path)
    fig, axes = plt.subplots(1, 2, sharey=True)
    fig.suptitle('Compounds with reproducible median correlation scores')
    fig.subplots_adjust(top=.80)
    axes = axes.flatten()
    fig.set_size_inches(18.7, 5.27)
    for idx, col in enumerate(['L1000', 'Cell painting']):
        sns.barplot(ax=axes[idx], x=df[col].index, y=df[col].values, palette=['gray'])
        axes[idx].set_title(df_cpd[col].name)
    [axes[i].set_ylabel("% Percent of compounds") for i in range(0, 1)]
    plt.savefig(os.path.join(path, file_name))
    plt.show()

##'#eb3434', '#eb7a34', '#ebae34', '#a45445', '#ea7862', '#eabe62'

len(df_cp_L1_melt['cpd'].unique())

1327

df_cpd = (df_cpd/len(df_cp_L1_melt['cpd'].unique())) * 100

df_cpd

	L1000	Cell painting
0.04 uM	25.470987	82.516956
0.12 uM	24.642050	77.694047
0.37 uM	27.505652	85.983421
1.11 uM	32.705350	91.183120
3.33 uM	33.986436	93.443858
10 uM	46.948003	98.040693

plot_no_of_cpd(df_cpd, 'L1000_cp_figures', 'no_of_reproducible_cpd.png')

Visualizing TAS, MAS and signature strength scores for L1000 vs Cell painting

def merge_L1000_CP_2(df_cp,df_L1):
    
    """Merge Cell painting and L1000 dataframes based on compounds and doses found in both dataframes"""
    
    cp_cpd = df_cp['cpd'].unique().tolist()
    L1_cpd = df_L1['cpd'].unique().tolist()
    all_cpds = [x for x in cp_cpd if x in L1_cpd]
    
    df_cp_scores =df_cp.loc[df_cp['cpd'].isin(all_cpds)].reset_index(drop=True)
    df_L1_scores = df_L1.loc[df_L1['cpd'].isin(all_cpds)].reset_index(drop=True)
    df_cp_scores.rename({'replicate_correlation':'median_scores_CP', 
                         'signature_strength':'Cell_painting_SS'}, axis = 1, inplace=True)
    df_L1_scores.rename({'replicate_correlation':'median_scores_L1000', 
                         'signature_strength':'L1000_SS'}, axis = 1, inplace=True)
    df_L1_cp = df_cp_scores.merge(df_L1_scores, on=['cpd', 'dose'])
    return df_L1_cp

df_L1_cp_all = merge_L1000_CP_2(df_cp_all,df_L1_all)

df_L1_cp_all[['No_of_reproducible_doses-Cell painting', 'No_of_reproducible_doses-L1000']] = \
df_L1_cp_med[['No_of_reproducible_doses-Cell painting', 'No_of_reproducible_doses-L1000']]

def plot_mas_vs_tas(df, title, L1_dmso_95pct, cp_dmso_95pct, path, file_name, hue = 'Cell painting', 
                    alp = 0.3, size =(50,300)):
    
    """Plot L1000 Transcriptional activity score (TAS) vs Cell Painting Morphological activity score (MAS)"""
    
    if not os.path.exists(path):
        os.mkdir(path)
    rel_plt = sns.relplot(data=df, x="MAS", y="TAS", col="dose", hue = 'No_of_reproducible_doses-' + hue,
                          size = 'No_of_reproducible_doses-' + hue, sizes= size, kind="scatter", 
                          palette=sns.diverging_palette(250, 30, l=65, center="dark", as_cmap=True),
                          col_wrap=3, height=5.5, aspect=1.3, alpha = alp)
    rel_plt.fig.suptitle(title)
    rel_plt.fig.subplots_adjust(top=.91)
    for idx in range(6):
        rel_plt.axes[idx].axhline(y = L1_dmso_95pct, ls='--', color = 'black', alpha = 0.8, label = '95th-DMSO TAS score')
        rel_plt.axes[idx].axvline(x = cp_dmso_95pct, ls='--', color = 'red', alpha = 0.8, label = '95th-DMSO MAS score')
    rel_leg = rel_plt._legend
    rel_leg.set_bbox_to_anchor([0.90, 0.51])
    plt.legend(bbox_to_anchor=(1.01, 0.495), loc='lower left')
    rel_plt.legend.set_title('Number of\nreproducible\ndoses\n' + '('+ hue +')')
    plt.savefig(os.path.join(path, file_name))
    plt.show()

plot_mas_vs_tas(df_L1_cp_all, "L1000 Transcriptional activity score (TAS) vs Cell Painting Morphological activity score (MAS)",
               L1_dmso_95pct, cp_dmso_95pct, 'L1000_cp_figures', 'TAS_vs_MAS_1a.png')

plot_mas_vs_tas(df_L1_cp_all, "L1000 Transcriptional activity score (TAS) vs Cell Painting Morphological activity score (MAS)",
               L1_dmso_95pct, cp_dmso_95pct, 'L1000_cp_figures', 'TAS_vs_MAS_1b.png', hue = 'L1000')

def plot_ss(df, title, path, file_name, hue = 'Cell painting', alp=0.3, size =(50,300)):
    
    """Plot L1000 vs Cell Painting signature strength"""
    if not os.path.exists(path):
        os.mkdir(path)
    rel_plt = sns.relplot(data=df, x='Cell_painting_SS', y='L1000_SS', col="dose", 
                          hue = 'No_of_reproducible_doses-' + hue, size = 'No_of_reproducible_doses-' + hue, 
                          sizes= size, kind="scatter", 
                          palette=sns.diverging_palette(250, 30, l=65, center="dark", as_cmap=True),
                          col_wrap=3, height=5.5, aspect=1.3, alpha = alp)
    rel_plt.fig.suptitle(title)
    rel_plt.fig.subplots_adjust(top=.91)
    rel_leg = rel_plt._legend
    rel_leg.set_bbox_to_anchor([0.90, 0.51])
    rel_plt.legend.set_title('Number of\nreproducible\ndoses\n' + '('+ hue +')')
    plt.savefig(os.path.join(path, file_name))
    plt.show()

plot_ss(df_L1_cp_all, "L1000 vs Cell Painting signature strength for compound replicates across all doses",
                'L1000_cp_figures', 'L1000_SS_vs_CP_SS_1a.png')

plot_ss(df_L1_cp_all, "L1000 vs Cell Painting signature strength for compound replicates across all doses",
                'L1000_cp_figures', 'L1000_SS_vs_CP_SS_1b.png', hue = 'L1000')

- Compounds with high MAS but low TAS (i.e. TAS below 0.3 and MAS > 0.7)

df_highmas_lowtas = df_L1_cp_all[(df_L1_cp_all['TAS'] < 0.3) & (df_L1_cp_all['MAS'] > 0.7)].copy()

plot_mas_vs_tas(df_highmas_lowtas, "Low L1000 TAS vs High Cell Painting MAS",
               L1_dmso_95pct, cp_dmso_95pct, 'L1000_cp_figures', 'Low_TAS_vs_High_MAS_1a.png', alp = 0.7)

plot_mas_vs_tas(df_highmas_lowtas, "Low L1000 TAS vs Cell Painting MAS",
               L1_dmso_95pct, cp_dmso_95pct, 'L1000_cp_figures', 'Low_TAS_vs_High_MAS_1b.png', 
                hue = 'L1000',alp = 0.8)

df_lowmas_hightas = df_L1_cp_all[(df_L1_cp_all['MAS'] < 0.2) & (df_L1_cp_all['TAS'] > 0.5)].copy()

df_lowmas_hightas

	cpd	dose	median_scores_CP	Cell_painting_SS	MAS	median_scores_L1000	L1000_SS	TAS	No_of_reproducible_doses-Cell painting	No_of_reproducible_doses-L1000
685	lasalocid	0.04 uM	0.075003	321.0	0.173371	0.623312	460.333333	0.541651	6	6
2003	l-ergothioneine	0.12 uM	0.058396	359.2	0.161824	0.640835	536.666667	0.593002	6	5
3330	l-ergothioneine	0.37 uM	0.063974	365.2	0.170786	0.569795	503.000000	0.541345	6	5
4105	amg-517	1.11 uM	0.059216	433.2	0.178956	0.620393	537.333333	0.583829	5	4
4625	iniparib	1.11 uM	0.047734	406.0	0.155547	0.461702	568.000000	0.517828	5	6
4657	l-ergothioneine	1.11 uM	0.049568	346.8	0.146495	0.598063	601.500000	0.606487	6	5
5183	tetrahydrouridine	1.11 uM	0.045648	338.2	0.138830	0.742995	505.000000	0.619397	5	3
5984	l-ergothioneine	3.33 uM	0.060725	395.6	0.173179	0.542090	526.666667	0.540299	6	5
6540	tofacitinib	3.33 uM	0.070028	411.0	0.189558	0.585557	537.666667	0.567377	6	3

- Compounds (i.e. p-values below 0.05) with reproducible median correlation values in 1.11 - 10uM doses

- Plot the MAS vs TAS;

- L1000 vs Cell painting Signature Strength

stat_cpds = df_stat_cp['cpd'].unique().tolist()

len(stat_cpds)

247

df_stat_all = df_L1_cp_all.loc[df_L1_cp_all['cpd'].isin(stat_cpds)].reset_index(drop=True)

plot_mas_vs_tas(df_stat_all, "L1000 TAS vs Cell Painting MAS for Compounds with reproducible median correlation values",
               L1_dmso_95pct, cp_dmso_95pct, 'L1000_cp_figures', 'stat_TAS_vs_MAS_1a.png', alp=0.5, size =(100,300))

plot_mas_vs_tas(df_stat_all, "L1000 TAS vs Cell Painting MAS for Compounds with reproducible median correlation values",
               L1_dmso_95pct, cp_dmso_95pct, 'L1000_cp_figures', 'stat_TAS_vs_MAS_1b.png',
                hue = 'L1000', alp=0.5, size =(100,300))

plot_ss(df_stat_all, "L1000 vs Cell Painting signature strength for Compounds with reproducible median correlation values",
                'L1000_cp_figures', 'stat_L1000_SS_vs_CP_SS_1a.png', alp=0.5, size =(100,300))

plot_ss(df_stat_all, "L1000 vs Cell Painting signature strength for Compounds with reproducible median correlation values",
        'L1000_cp_figures', 'stat_L1000_SS_vs_CP_SS_1b.png', 
        hue = 'L1000', alp=0.5, size =(100,300))